ALEXANDER GROSU AND FRED LANDMAN STRANGE RELATIVES OF THE THIRD KIND* In this paper, we argue that there are more kinds of relative clause constructions between the linguistic heaven and earth than are dreamed of in the classical lore, which distinguishes just restrictive relative clauses and appositives. We start with degree relatives. Degree, or amount, relatives show restrictions in the relativizers they allow, in the determiners that can combine with them, and in their stacking possibilities. To account for these facts, we propose an analysis with two central, and novel, features: First, we argue that the standard notion of degree (a number on a measuring scale) needs to be replaced by a notion of structured degree, which keeps track of the object measured. Second, we argue that at the CP-level of degree relatives an operation of (degree) maximalization takes place. We show that the observed facts concerning degree relatives follow from these assumptions. We then broaden the discussion to other relative clause constructions. We propose that the operation of maximalization takes place in relative clauses when the head noun is semantically interpreted CP- internally, while syntactically the CP is part of a DP that also contains CP-external material. Based on this, we argue that degree relatives form part of a linguistically coherent class of relative clause constructions – we call them maximalizing relatives – which all show restrictions similar to those observed for degree relatives, and which differ semantically (and often also syntactically) both from restrictive relative clauses and from appositives. We discuss free relatives, internally-headed relatives, and cor- relatives. 1. INTRODUCTION Much of the traditional and generative literature has assumed two semantic types of relative clause constructions: restrictives and appositives. Following Partee (1973), it is standardly assumed that restrictive relative clauses denote sets which semantically combine with their head noun through set inter- * Versions of this paper were presented by Fred at the Israel Semantics Circle (Spring 1995), the 11th Annual Meeting of the Israel Association for Theoretical Linguistics (June 1995), and the 10th Amsterdam Colloquium (December 1995); and by Alex at the 1st Paris Conference on Syntax and Semantics (October 1995), the Conference on (Preferentially) Non- Lexical Semantics (Paris, June 1996), the Workshop on Discourse of the Israel Association for Theoretical Linguistics (Spring 1996), and the 23rd Incontro di Grammatica Generativa (Pisa, 1997). We thank the audiences of these presentations for their helpful comments and discussions. In addition, many people have helped us sharpen our theoretical discussion, or answered intricate data questions. We would like to thank especially Artemis Alexiadou, Greg Carlson, Gennaro Chierchia, Veneeta Dayal, Mariana Grosu, Gabriella Hermon, Nirit Kadmon, Hotze Rullmann, Susan Rothstein, Janis Williamson, and two anonymous referees. Finally, for remaining mistakes, the authors have decided not to blame each other, but to look for an appropriate scapegoat. Natural Language Semantics 6: 125–170, 1998. 1998 Kluwer Academic Publishers. Printed in the Netherlands. NALS Art No. 176 PIPS. NO. 156629 126 ALEXANDER GROSU AND FRED LANDMAN section. Since more than one set can intersect with the same head noun, restrictive relative clauses can stack. As Sells (1985) shows, appositive relative clauses contain an element that stands in a discourse anaphora relation to the NP they modify. Since more than one relative can stand in a discourse anaphora relation to the same NP, appositive relative clauses can stack too. We are concerned in this paper with relative clauses that for a variety of reasons do not fit either of these molds. The kinds of construction we will be looking at are: – degree relatives (discussed in Carlson 1977b and Heim 1987) – realis and irrealis free relatives (realis: e.g. Jacobson 1988) – internally headed relatives (Basilico 1996) – correlatives (Dayal 1991a, b, 1996). As we will see, one diagnostic that distinguishes these constructions from both restrictive and appositive relative clauses is that they do not allow stacking (excepting cases with a special semantics, such as corrections, where a second relative clause corrects or amends an earlier one). What we will argue is that the traditional emphasis on appositives and restrictives gives a rather lopsided picture of the full class of relative clause constructions. We propose that a more satisfactory classification of relative clauses is in terms of sortal-internal vs. sortal-external relatives. This dichotomy takes as a criterion whether the relative construction’s sortal (which is the common noun, or NP, if nominal constructions are DPs) is semantically construed as outside or inside the construction’s CP. In certain kinds of relative clauses, like correlatives, there can actually be on the surface two tokens of a sortal, one CP-internal and one CP-external. But the situation is essentially the same here: the relevant criterion is which of the two tokens is the one that contributes to the semantics; the other one is, in essence, semantically vacuous. The classical picture is lopsided because both appositives and restrictives fall inside the class of sortal-external relatives. This makes everything else a “strange relative of the third kind.” But it should be clear that this name only indicates the relatively late appearance of non-appositive, non-restric- tive relatives on the linguistic horizon. If we are right in claiming that the main semantic dichotomy within relatives is in terms of the sortal- internal/external distinction, then there is nothing strange about the internal ones, nor are they properly of a third kind. Rather, we think of relatives as taking positions in a spectrum of the following sort: Simplex XPs – Appositives – Restrictives – Maximalizers – Simplex CPs 123 4 5 NALS Art No. 176 PIPS. NO. 156629 STRANGE RELATIVES OF THE THIRD KIND 127 Relative clauses form a paradigm in which various intermediate options are filled in between two extremes. On the one extreme (1), there is no relative to make a contribution to the construction whatsoever: these are simplex XPs which lack a relative clause altogether. On the other extreme (5), there is no contribution to the construction besides the relative: in this case the entire construction is a bare CP, completely lacking external material. We will argue that this situation is represented by the irrealis free relatives. In the center of the paradigm, we find restrictives. These are construc- tions where the head noun and the relative provide, by and large, an equal contribution to the entire construction, shown by the fact that they combine through intersection, which is a symmetric operation. Appositives are on the left side of the paradigm: the CP-external material makes the main contribution to the construction; since the appositive is related to the CP- external material as a discourse anaphor, its contribution to the construction is indirect and mediated through the discourse level (see Sells 1985 for details). Between the center and the right extreme (under 3), we find relative clause constructions where the contribution of the CP-external material is reduced to the minimum consistent with the CP’s internal makeup. This is where we find the four other constructions that we will discuss. We will argue that part of the semantics of these constructions is a CP-internal maximalization operation. Whatever CP-external material is syntactically present is either interpreted CP-internally or by and large predictable from the semantic interpretation of the CP after maximalization. As indicated at the beginning of this paper, the semantic operations involved in appositives and restrictives easily generalize to n-place opera- tions (i.e., it is in principle as easy to connect n discourse anaphors to an NP as it is to connect one; similarly, intersecting a noun with n relatives is as easy as restricting it with one). This is the reason that these relatives stack. On the right side of the paradigm, what CP-external material there is (semantically) is predicted from the CP-internal semantics. This means that the sortal and cardinality properties are fixed CP-internally. We assume that it is not possible to independently fix these properties more than once for the same construction. Hence the relatives on the right side do not stack. The structure of this paper is as follows. We will first discuss the syntax and semantics of degree relatives as a paradigm case of maximalizing rel- atives. After that we will discuss realis and irrealis free relatives, internally headed relatives, and correlatives. NALS Art No. 176 PIPS. NO. 156629 128 ALEXANDER GROSU AND FRED LANDMAN 2. DEGREE RELATIVES 2.1. Carlson and Heim Degree relatives were discussed in Carlson (1977b) (where they are called ‘amount relatives’) and Heim (1987). Carlson draws attention to the following facts concerning the interac- tion between relativization and there-insertion contexts. If the relative clause contains a there-insertion context and the relativization gap is in the position which is open to the definiteness effect, the relative clause is OK with the relativizer that or with the empty relativizer Ø, but not with the rela- tivizer which: (1) a. I took with me the three books that/Ø there were __ on the table. b.#I took with me the three books which there were __ on the table. While their accounts are couched in different frameworks, Carlson and Heim give in essence the same explanation for the infelicity of (1b) (we follow Heim’s account): – The gap of relativization with relativizer which is filled by an indi- vidual variable. – Individual variables count as strong NPs. – This means that (1b) contains a strong NP in the position which is open to the definiteness effect, hence (1b) is infelicitous. This account of the infelicity of (1b) is neutral with respect to the proper account of the definiteness effect, but, of course, it relies on the assump- tion that bound individual variables count as strong.